One API to manage all your AI
🔒 VPS onlyThe industrial LLM gateway that unifies all your models, measures their performance, and optimizes prompts and costs. Installed and supervised by DINAO, your inferences and feedbacks remain on our French servers.
What is TensorZero?
TensorZero is an open-source platform that unifies in a single product an LLM gateway, observability, evaluation, optimization, and experimentation. It exposes a single OpenAI-compatible API providing access to all major model providers, and stores every inference and feedback to subsequently optimize prompts, models, and inference strategies.
The gateway is written in Rust to achieve industrial performance: less than one millisecond of latency at p99 and several thousand requests per second. Intelligent routing, automatic fallbacks, retries, cache, rate limiting, cost tracking, and A/B testing are natively integrated, as well as support for structured outputs, tool use, embeddings, multimodal, and batch processing.
On the data side, observability relies on ClickHouse (recommended for high volume) or PostgreSQL (simplest configuration), and deployment is done via Docker or Kubernetes with GitOps-compatible configuration.
Host TensorZero at DINAO
Resource tiers compatible with TensorZero prerequisites (minimum 1 vCPU (gateway 0,5 vCPU + store) / 512 Mo (gateway) + 1 Go (store) / 5 à 10 Go). Hosted in France, fully managed.
Application available on a dedicated VPS
This application requires privileged access to the host machine, incompatible with managed shared hosting. It is installed on a dedicated VPS server where you keep full control — managed by DINAO.
Technical reason: bind-mount d'un chemin hôte (volumes nommés uniquement en managé).
Discover the Docker VPS →Requirements for TensorZero: 1 vCPU (gateway 0,5 vCPU + store), 512 Mo (gateway) + 1 Go (store), 5 à 10 Go. Our Docker VPS at €11 excl. VAT per month covers them.
This application uses AI
The container hosts the application, not the AI engine (which requires dedicated GPUs) : inference is handled externally, with your own provider key. Prioritize a sovereign engine — Mistral AI, NumSpot, Scaleway, or OVHcloud AI Endpoints (France, GDPR) ; an international provider (OpenAI, Anthropic…) only if a specific capability requires it. Inference subscriptions are not included in hosting.
Technical details
You might be wondering…
Does TensorZero host the models themselves?
No. TensorZero is a gateway: it routes your requests to providers (Anthropic, OpenAI, Mistral…) or to a local inference server. It is what unifies access, observability, and optimization — no GPU required on the instance.
Do I need to rewrite my application code?
No. The API is OpenAI-compatible: simply point your SDK's base_url to your DINAO instance. You can then change models or providers without touching the code.
Where are the data hosted?
On DINAO infrastructure in France, in one of the available data centers. Your inference logs, prompts, and feedbacks do not leave the territory.
Are technical skills required?
DINAO manages installation, the observability backend (ClickHouse/PostgreSQL), security, and updates. Your technical team focuses on application integration and result analysis.
Can I change plans or export my data?
Yes. You can upgrade or downgrade at any time, and your inferences remain exportable — no proprietary lock-in.
